]> git.sesse.net Git - ffmpeg/blob - libavcodec/mpegvideo.c
Merge commit '28a8b5413b64b831dfb8650208bccd8b78360484'
[ffmpeg] / libavcodec / mpegvideo.c
1 /*
2  * The simplest mpeg encoder (well, it was the simplest!)
3  * Copyright (c) 2000,2001 Fabrice Bellard
4  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5  *
6  * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
7  *
8  * This file is part of FFmpeg.
9  *
10  * FFmpeg is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU Lesser General Public
12  * License as published by the Free Software Foundation; either
13  * version 2.1 of the License, or (at your option) any later version.
14  *
15  * FFmpeg is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * Lesser General Public License for more details.
19  *
20  * You should have received a copy of the GNU Lesser General Public
21  * License along with FFmpeg; if not, write to the Free Software
22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23  */
24
25 /**
26  * @file
27  * The simplest mpeg encoder (well, it was the simplest!).
28  */
29
30 #include "libavutil/attributes.h"
31 #include "libavutil/avassert.h"
32 #include "libavutil/imgutils.h"
33 #include "libavutil/internal.h"
34 #include "libavutil/motion_vector.h"
35 #include "libavutil/timer.h"
36 #include "avcodec.h"
37 #include "blockdsp.h"
38 #include "h264chroma.h"
39 #include "idctdsp.h"
40 #include "internal.h"
41 #include "mathops.h"
42 #include "mpeg_er.h"
43 #include "mpegutils.h"
44 #include "mpegvideo.h"
45 #include "mpegvideodata.h"
46 #include "mjpegenc.h"
47 #include "msmpeg4.h"
48 #include "qpeldsp.h"
49 #include "thread.h"
50 #include "wmv2.h"
51 #include <limits.h>
52
53 static void dct_unquantize_mpeg1_intra_c(MpegEncContext *s,
54                                    int16_t *block, int n, int qscale)
55 {
56     int i, level, nCoeffs;
57     const uint16_t *quant_matrix;
58
59     nCoeffs= s->block_last_index[n];
60
61     block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
62     /* XXX: only MPEG-1 */
63     quant_matrix = s->intra_matrix;
64     for(i=1;i<=nCoeffs;i++) {
65         int j= s->intra_scantable.permutated[i];
66         level = block[j];
67         if (level) {
68             if (level < 0) {
69                 level = -level;
70                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
71                 level = (level - 1) | 1;
72                 level = -level;
73             } else {
74                 level = (int)(level * qscale * quant_matrix[j]) >> 3;
75                 level = (level - 1) | 1;
76             }
77             block[j] = level;
78         }
79     }
80 }
81
82 static void dct_unquantize_mpeg1_inter_c(MpegEncContext *s,
83                                    int16_t *block, int n, int qscale)
84 {
85     int i, level, nCoeffs;
86     const uint16_t *quant_matrix;
87
88     nCoeffs= s->block_last_index[n];
89
90     quant_matrix = s->inter_matrix;
91     for(i=0; i<=nCoeffs; i++) {
92         int j= s->intra_scantable.permutated[i];
93         level = block[j];
94         if (level) {
95             if (level < 0) {
96                 level = -level;
97                 level = (((level << 1) + 1) * qscale *
98                          ((int) (quant_matrix[j]))) >> 4;
99                 level = (level - 1) | 1;
100                 level = -level;
101             } else {
102                 level = (((level << 1) + 1) * qscale *
103                          ((int) (quant_matrix[j]))) >> 4;
104                 level = (level - 1) | 1;
105             }
106             block[j] = level;
107         }
108     }
109 }
110
111 static void dct_unquantize_mpeg2_intra_c(MpegEncContext *s,
112                                    int16_t *block, int n, int qscale)
113 {
114     int i, level, nCoeffs;
115     const uint16_t *quant_matrix;
116
117     if (s->q_scale_type) qscale = ff_mpeg2_non_linear_qscale[qscale];
118     else                 qscale <<= 1;
119
120     if(s->alternate_scan) nCoeffs= 63;
121     else nCoeffs= s->block_last_index[n];
122
123     block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
124     quant_matrix = s->intra_matrix;
125     for(i=1;i<=nCoeffs;i++) {
126         int j= s->intra_scantable.permutated[i];
127         level = block[j];
128         if (level) {
129             if (level < 0) {
130                 level = -level;
131                 level = (int)(level * qscale * quant_matrix[j]) >> 4;
132                 level = -level;
133             } else {
134                 level = (int)(level * qscale * quant_matrix[j]) >> 4;
135             }
136             block[j] = level;
137         }
138     }
139 }
140
141 static void dct_unquantize_mpeg2_intra_bitexact(MpegEncContext *s,
142                                    int16_t *block, int n, int qscale)
143 {
144     int i, level, nCoeffs;
145     const uint16_t *quant_matrix;
146     int sum=-1;
147
148     if (s->q_scale_type) qscale = ff_mpeg2_non_linear_qscale[qscale];
149     else                 qscale <<= 1;
150
151     if(s->alternate_scan) nCoeffs= 63;
152     else nCoeffs= s->block_last_index[n];
153
154     block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
155     sum += block[0];
156     quant_matrix = s->intra_matrix;
157     for(i=1;i<=nCoeffs;i++) {
158         int j= s->intra_scantable.permutated[i];
159         level = block[j];
160         if (level) {
161             if (level < 0) {
162                 level = -level;
163                 level = (int)(level * qscale * quant_matrix[j]) >> 4;
164                 level = -level;
165             } else {
166                 level = (int)(level * qscale * quant_matrix[j]) >> 4;
167             }
168             block[j] = level;
169             sum+=level;
170         }
171     }
172     block[63]^=sum&1;
173 }
174
175 static void dct_unquantize_mpeg2_inter_c(MpegEncContext *s,
176                                    int16_t *block, int n, int qscale)
177 {
178     int i, level, nCoeffs;
179     const uint16_t *quant_matrix;
180     int sum=-1;
181
182     if (s->q_scale_type) qscale = ff_mpeg2_non_linear_qscale[qscale];
183     else                 qscale <<= 1;
184
185     if(s->alternate_scan) nCoeffs= 63;
186     else nCoeffs= s->block_last_index[n];
187
188     quant_matrix = s->inter_matrix;
189     for(i=0; i<=nCoeffs; i++) {
190         int j= s->intra_scantable.permutated[i];
191         level = block[j];
192         if (level) {
193             if (level < 0) {
194                 level = -level;
195                 level = (((level << 1) + 1) * qscale *
196                          ((int) (quant_matrix[j]))) >> 5;
197                 level = -level;
198             } else {
199                 level = (((level << 1) + 1) * qscale *
200                          ((int) (quant_matrix[j]))) >> 5;
201             }
202             block[j] = level;
203             sum+=level;
204         }
205     }
206     block[63]^=sum&1;
207 }
208
209 static void dct_unquantize_h263_intra_c(MpegEncContext *s,
210                                   int16_t *block, int n, int qscale)
211 {
212     int i, level, qmul, qadd;
213     int nCoeffs;
214
215     av_assert2(s->block_last_index[n]>=0 || s->h263_aic);
216
217     qmul = qscale << 1;
218
219     if (!s->h263_aic) {
220         block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
221         qadd = (qscale - 1) | 1;
222     }else{
223         qadd = 0;
224     }
225     if(s->ac_pred)
226         nCoeffs=63;
227     else
228         nCoeffs= s->intra_scantable.raster_end[ s->block_last_index[n] ];
229
230     for(i=1; i<=nCoeffs; i++) {
231         level = block[i];
232         if (level) {
233             if (level < 0) {
234                 level = level * qmul - qadd;
235             } else {
236                 level = level * qmul + qadd;
237             }
238             block[i] = level;
239         }
240     }
241 }
242
243 static void dct_unquantize_h263_inter_c(MpegEncContext *s,
244                                   int16_t *block, int n, int qscale)
245 {
246     int i, level, qmul, qadd;
247     int nCoeffs;
248
249     av_assert2(s->block_last_index[n]>=0);
250
251     qadd = (qscale - 1) | 1;
252     qmul = qscale << 1;
253
254     nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
255
256     for(i=0; i<=nCoeffs; i++) {
257         level = block[i];
258         if (level) {
259             if (level < 0) {
260                 level = level * qmul - qadd;
261             } else {
262                 level = level * qmul + qadd;
263             }
264             block[i] = level;
265         }
266     }
267 }
268
269
270 static void gray16(uint8_t *dst, const uint8_t *src, ptrdiff_t linesize, int h)
271 {
272     while(h--)
273         memset(dst + h*linesize, 128, 16);
274 }
275
276 static void gray8(uint8_t *dst, const uint8_t *src, ptrdiff_t linesize, int h)
277 {
278     while(h--)
279         memset(dst + h*linesize, 128, 8);
280 }
281
282 /* init common dct for both encoder and decoder */
283 static av_cold int dct_init(MpegEncContext *s)
284 {
285     ff_blockdsp_init(&s->bdsp, s->avctx);
286     ff_h264chroma_init(&s->h264chroma, 8); //for lowres
287     ff_hpeldsp_init(&s->hdsp, s->avctx->flags);
288     ff_mpegvideodsp_init(&s->mdsp);
289     ff_videodsp_init(&s->vdsp, s->avctx->bits_per_raw_sample);
290
291     if (s->avctx->debug & FF_DEBUG_NOMC) {
292         int i;
293         for (i=0; i<4; i++) {
294             s->hdsp.avg_pixels_tab[0][i] = gray16;
295             s->hdsp.put_pixels_tab[0][i] = gray16;
296             s->hdsp.put_no_rnd_pixels_tab[0][i] = gray16;
297
298             s->hdsp.avg_pixels_tab[1][i] = gray8;
299             s->hdsp.put_pixels_tab[1][i] = gray8;
300             s->hdsp.put_no_rnd_pixels_tab[1][i] = gray8;
301         }
302     }
303
304     s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_c;
305     s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_c;
306     s->dct_unquantize_mpeg1_intra = dct_unquantize_mpeg1_intra_c;
307     s->dct_unquantize_mpeg1_inter = dct_unquantize_mpeg1_inter_c;
308     s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_c;
309     if (s->avctx->flags & AV_CODEC_FLAG_BITEXACT)
310         s->dct_unquantize_mpeg2_intra = dct_unquantize_mpeg2_intra_bitexact;
311     s->dct_unquantize_mpeg2_inter = dct_unquantize_mpeg2_inter_c;
312
313     if (HAVE_INTRINSICS_NEON)
314         ff_mpv_common_init_neon(s);
315
316     if (ARCH_ALPHA)
317         ff_mpv_common_init_axp(s);
318     if (ARCH_ARM)
319         ff_mpv_common_init_arm(s);
320     if (ARCH_PPC)
321         ff_mpv_common_init_ppc(s);
322     if (ARCH_X86)
323         ff_mpv_common_init_x86(s);
324     if (ARCH_MIPS)
325         ff_mpv_common_init_mips(s);
326
327     return 0;
328 }
329
330 av_cold void ff_mpv_idct_init(MpegEncContext *s)
331 {
332     if (s->codec_id == AV_CODEC_ID_MPEG4)
333         s->idsp.mpeg4_studio_profile = s->studio_profile;
334     ff_idctdsp_init(&s->idsp, s->avctx);
335
336     /* load & permutate scantables
337      * note: only wmv uses different ones
338      */
339     if (s->alternate_scan) {
340         ff_init_scantable(s->idsp.idct_permutation, &s->inter_scantable, ff_alternate_vertical_scan);
341         ff_init_scantable(s->idsp.idct_permutation, &s->intra_scantable, ff_alternate_vertical_scan);
342     } else {
343         ff_init_scantable(s->idsp.idct_permutation, &s->inter_scantable, ff_zigzag_direct);
344         ff_init_scantable(s->idsp.idct_permutation, &s->intra_scantable, ff_zigzag_direct);
345     }
346     ff_init_scantable(s->idsp.idct_permutation, &s->intra_h_scantable, ff_alternate_horizontal_scan);
347     ff_init_scantable(s->idsp.idct_permutation, &s->intra_v_scantable, ff_alternate_vertical_scan);
348 }
349
350 static int alloc_picture(MpegEncContext *s, Picture *pic, int shared)
351 {
352     return ff_alloc_picture(s->avctx, pic, &s->me, &s->sc, shared, 0,
353                             s->chroma_x_shift, s->chroma_y_shift, s->out_format,
354                             s->mb_stride, s->mb_width, s->mb_height, s->b8_stride,
355                             &s->linesize, &s->uvlinesize);
356 }
357
358 static int init_duplicate_context(MpegEncContext *s)
359 {
360     int y_size = s->b8_stride * (2 * s->mb_height + 1);
361     int c_size = s->mb_stride * (s->mb_height + 1);
362     int yc_size = y_size + 2 * c_size;
363     int i;
364
365     if (s->mb_height & 1)
366         yc_size += 2*s->b8_stride + 2*s->mb_stride;
367
368     s->sc.edge_emu_buffer =
369     s->me.scratchpad   =
370     s->me.temp         =
371     s->sc.rd_scratchpad   =
372     s->sc.b_scratchpad    =
373     s->sc.obmc_scratchpad = NULL;
374
375     if (s->encoding) {
376         FF_ALLOCZ_OR_GOTO(s->avctx, s->me.map,
377                           ME_MAP_SIZE * sizeof(uint32_t), fail)
378         FF_ALLOCZ_OR_GOTO(s->avctx, s->me.score_map,
379                           ME_MAP_SIZE * sizeof(uint32_t), fail)
380         if (s->noise_reduction) {
381             FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_error_sum,
382                               2 * 64 * sizeof(int), fail)
383         }
384     }
385     FF_ALLOCZ_OR_GOTO(s->avctx, s->blocks, 64 * 12 * 2 * sizeof(int16_t), fail)
386     s->block = s->blocks[0];
387
388     for (i = 0; i < 12; i++) {
389         s->pblocks[i] = &s->block[i];
390     }
391
392     FF_ALLOCZ_OR_GOTO(s->avctx, s->block32, sizeof(*s->block32), fail)
393     s->dpcm_direction = 0;
394     FF_ALLOCZ_OR_GOTO(s->avctx, s->dpcm_macroblock, sizeof(*s->dpcm_macroblock), fail)
395
396     if (s->avctx->codec_tag == AV_RL32("VCR2")) {
397         // exchange uv
398         FFSWAP(void *, s->pblocks[4], s->pblocks[5]);
399     }
400
401     if (s->out_format == FMT_H263) {
402         /* ac values */
403         FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_val_base,
404                           yc_size * sizeof(int16_t) * 16, fail);
405         s->ac_val[0] = s->ac_val_base + s->b8_stride + 1;
406         s->ac_val[1] = s->ac_val_base + y_size + s->mb_stride + 1;
407         s->ac_val[2] = s->ac_val[1] + c_size;
408     }
409
410     return 0;
411 fail:
412     return -1; // free() through ff_mpv_common_end()
413 }
414
415 static void free_duplicate_context(MpegEncContext *s)
416 {
417     if (!s)
418         return;
419
420     av_freep(&s->sc.edge_emu_buffer);
421     av_freep(&s->me.scratchpad);
422     s->me.temp =
423     s->sc.rd_scratchpad =
424     s->sc.b_scratchpad =
425     s->sc.obmc_scratchpad = NULL;
426
427     av_freep(&s->dct_error_sum);
428     av_freep(&s->me.map);
429     av_freep(&s->me.score_map);
430     av_freep(&s->blocks);
431     av_freep(&s->block32);
432     av_freep(&s->dpcm_macroblock);
433     av_freep(&s->ac_val_base);
434     s->block = NULL;
435 }
436
437 static void backup_duplicate_context(MpegEncContext *bak, MpegEncContext *src)
438 {
439 #define COPY(a) bak->a = src->a
440     COPY(sc.edge_emu_buffer);
441     COPY(me.scratchpad);
442     COPY(me.temp);
443     COPY(sc.rd_scratchpad);
444     COPY(sc.b_scratchpad);
445     COPY(sc.obmc_scratchpad);
446     COPY(me.map);
447     COPY(me.score_map);
448     COPY(blocks);
449     COPY(block);
450     COPY(block32);
451     COPY(dpcm_macroblock);
452     COPY(dpcm_direction);
453     COPY(start_mb_y);
454     COPY(end_mb_y);
455     COPY(me.map_generation);
456     COPY(pb);
457     COPY(dct_error_sum);
458     COPY(dct_count[0]);
459     COPY(dct_count[1]);
460     COPY(ac_val_base);
461     COPY(ac_val[0]);
462     COPY(ac_val[1]);
463     COPY(ac_val[2]);
464 #undef COPY
465 }
466
467 int ff_update_duplicate_context(MpegEncContext *dst, MpegEncContext *src)
468 {
469     MpegEncContext bak;
470     int i, ret;
471     // FIXME copy only needed parts
472     // START_TIMER
473     backup_duplicate_context(&bak, dst);
474     memcpy(dst, src, sizeof(MpegEncContext));
475     backup_duplicate_context(dst, &bak);
476     for (i = 0; i < 12; i++) {
477         dst->pblocks[i] = &dst->block[i];
478     }
479     if (dst->avctx->codec_tag == AV_RL32("VCR2")) {
480         // exchange uv
481         FFSWAP(void *, dst->pblocks[4], dst->pblocks[5]);
482     }
483     if (!dst->sc.edge_emu_buffer &&
484         (ret = ff_mpeg_framesize_alloc(dst->avctx, &dst->me,
485                                        &dst->sc, dst->linesize)) < 0) {
486         av_log(dst->avctx, AV_LOG_ERROR, "failed to allocate context "
487                "scratch buffers.\n");
488         return ret;
489     }
490     // STOP_TIMER("update_duplicate_context")
491     // about 10k cycles / 0.01 sec for  1000frames on 1ghz with 2 threads
492     return 0;
493 }
494
495 int ff_mpeg_update_thread_context(AVCodecContext *dst,
496                                   const AVCodecContext *src)
497 {
498     int i, ret;
499     MpegEncContext *s = dst->priv_data, *s1 = src->priv_data;
500
501     if (dst == src)
502         return 0;
503
504     av_assert0(s != s1);
505
506     // FIXME can parameters change on I-frames?
507     // in that case dst may need a reinit
508     if (!s->context_initialized) {
509         int err;
510         memcpy(s, s1, sizeof(MpegEncContext));
511
512         s->avctx                 = dst;
513         s->bitstream_buffer      = NULL;
514         s->bitstream_buffer_size = s->allocated_bitstream_buffer_size = 0;
515
516         if (s1->context_initialized){
517 //             s->picture_range_start  += MAX_PICTURE_COUNT;
518 //             s->picture_range_end    += MAX_PICTURE_COUNT;
519             ff_mpv_idct_init(s);
520             if((err = ff_mpv_common_init(s)) < 0){
521                 memset(s, 0, sizeof(MpegEncContext));
522                 s->avctx = dst;
523                 return err;
524             }
525         }
526     }
527
528     if (s->height != s1->height || s->width != s1->width || s->context_reinit) {
529         s->context_reinit = 0;
530         s->height = s1->height;
531         s->width  = s1->width;
532         if ((ret = ff_mpv_common_frame_size_change(s)) < 0)
533             return ret;
534     }
535
536     s->avctx->coded_height  = s1->avctx->coded_height;
537     s->avctx->coded_width   = s1->avctx->coded_width;
538     s->avctx->width         = s1->avctx->width;
539     s->avctx->height        = s1->avctx->height;
540
541     s->quarter_sample       = s1->quarter_sample;
542
543     s->coded_picture_number = s1->coded_picture_number;
544     s->picture_number       = s1->picture_number;
545
546     av_assert0(!s->picture || s->picture != s1->picture);
547     if(s->picture)
548     for (i = 0; i < MAX_PICTURE_COUNT; i++) {
549         ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
550         if (s1->picture && s1->picture[i].f->buf[0] &&
551             (ret = ff_mpeg_ref_picture(s->avctx, &s->picture[i], &s1->picture[i])) < 0)
552             return ret;
553     }
554
555 #define UPDATE_PICTURE(pic)\
556 do {\
557     ff_mpeg_unref_picture(s->avctx, &s->pic);\
558     if (s1->pic.f && s1->pic.f->buf[0])\
559         ret = ff_mpeg_ref_picture(s->avctx, &s->pic, &s1->pic);\
560     else\
561         ret = ff_update_picture_tables(&s->pic, &s1->pic);\
562     if (ret < 0)\
563         return ret;\
564 } while (0)
565
566     UPDATE_PICTURE(current_picture);
567     UPDATE_PICTURE(last_picture);
568     UPDATE_PICTURE(next_picture);
569
570 #define REBASE_PICTURE(pic, new_ctx, old_ctx)                                 \
571     ((pic && pic >= old_ctx->picture &&                                       \
572       pic < old_ctx->picture + MAX_PICTURE_COUNT) ?                           \
573         &new_ctx->picture[pic - old_ctx->picture] : NULL)
574
575     s->last_picture_ptr    = REBASE_PICTURE(s1->last_picture_ptr,    s, s1);
576     s->current_picture_ptr = REBASE_PICTURE(s1->current_picture_ptr, s, s1);
577     s->next_picture_ptr    = REBASE_PICTURE(s1->next_picture_ptr,    s, s1);
578
579     // Error/bug resilience
580     s->next_p_frame_damaged = s1->next_p_frame_damaged;
581     s->workaround_bugs      = s1->workaround_bugs;
582     s->padding_bug_score    = s1->padding_bug_score;
583
584     // MPEG-4 timing info
585     memcpy(&s->last_time_base, &s1->last_time_base,
586            (char *) &s1->pb_field_time + sizeof(s1->pb_field_time) -
587            (char *) &s1->last_time_base);
588
589     // B-frame info
590     s->max_b_frames = s1->max_b_frames;
591     s->low_delay    = s1->low_delay;
592     s->droppable    = s1->droppable;
593
594     // DivX handling (doesn't work)
595     s->divx_packed  = s1->divx_packed;
596
597     if (s1->bitstream_buffer) {
598         if (s1->bitstream_buffer_size +
599             AV_INPUT_BUFFER_PADDING_SIZE > s->allocated_bitstream_buffer_size) {
600             av_fast_malloc(&s->bitstream_buffer,
601                            &s->allocated_bitstream_buffer_size,
602                            s1->allocated_bitstream_buffer_size);
603             if (!s->bitstream_buffer) {
604                 s->bitstream_buffer_size = 0;
605                 return AVERROR(ENOMEM);
606             }
607         }
608         s->bitstream_buffer_size = s1->bitstream_buffer_size;
609         memcpy(s->bitstream_buffer, s1->bitstream_buffer,
610                s1->bitstream_buffer_size);
611         memset(s->bitstream_buffer + s->bitstream_buffer_size, 0,
612                AV_INPUT_BUFFER_PADDING_SIZE);
613     }
614
615     // linesize-dependent scratch buffer allocation
616     if (!s->sc.edge_emu_buffer)
617         if (s1->linesize) {
618             if (ff_mpeg_framesize_alloc(s->avctx, &s->me,
619                                         &s->sc, s1->linesize) < 0) {
620                 av_log(s->avctx, AV_LOG_ERROR, "Failed to allocate context "
621                        "scratch buffers.\n");
622                 return AVERROR(ENOMEM);
623             }
624         } else {
625             av_log(s->avctx, AV_LOG_ERROR, "Context scratch buffers could not "
626                    "be allocated due to unknown size.\n");
627         }
628
629     // MPEG-2/interlacing info
630     memcpy(&s->progressive_sequence, &s1->progressive_sequence,
631            (char *) &s1->rtp_mode - (char *) &s1->progressive_sequence);
632
633     if (!s1->first_field) {
634         s->last_pict_type = s1->pict_type;
635         if (s1->current_picture_ptr)
636             s->last_lambda_for[s1->pict_type] = s1->current_picture_ptr->f->quality;
637     }
638
639     return 0;
640 }
641
642 /**
643  * Set the given MpegEncContext to common defaults
644  * (same for encoding and decoding).
645  * The changed fields will not depend upon the
646  * prior state of the MpegEncContext.
647  */
648 void ff_mpv_common_defaults(MpegEncContext *s)
649 {
650     s->y_dc_scale_table      =
651     s->c_dc_scale_table      = ff_mpeg1_dc_scale_table;
652     s->chroma_qscale_table   = ff_default_chroma_qscale_table;
653     s->progressive_frame     = 1;
654     s->progressive_sequence  = 1;
655     s->picture_structure     = PICT_FRAME;
656
657     s->coded_picture_number  = 0;
658     s->picture_number        = 0;
659
660     s->f_code                = 1;
661     s->b_code                = 1;
662
663     s->slice_context_count   = 1;
664 }
665
666 /**
667  * Set the given MpegEncContext to defaults for decoding.
668  * the changed fields will not depend upon
669  * the prior state of the MpegEncContext.
670  */
671 void ff_mpv_decode_defaults(MpegEncContext *s)
672 {
673     ff_mpv_common_defaults(s);
674 }
675
676 void ff_mpv_decode_init(MpegEncContext *s, AVCodecContext *avctx)
677 {
678     s->avctx           = avctx;
679     s->width           = avctx->coded_width;
680     s->height          = avctx->coded_height;
681     s->codec_id        = avctx->codec->id;
682     s->workaround_bugs = avctx->workaround_bugs;
683
684     /* convert fourcc to upper case */
685     s->codec_tag          = avpriv_toupper4(avctx->codec_tag);
686 }
687
688 /**
689  * Initialize and allocates MpegEncContext fields dependent on the resolution.
690  */
691 static int init_context_frame(MpegEncContext *s)
692 {
693     int y_size, c_size, yc_size, i, mb_array_size, mv_table_size, x, y;
694
695     s->mb_width   = (s->width + 15) / 16;
696     s->mb_stride  = s->mb_width + 1;
697     s->b8_stride  = s->mb_width * 2 + 1;
698     mb_array_size = s->mb_height * s->mb_stride;
699     mv_table_size = (s->mb_height + 2) * s->mb_stride + 1;
700
701     /* set default edge pos, will be overridden
702      * in decode_header if needed */
703     s->h_edge_pos = s->mb_width * 16;
704     s->v_edge_pos = s->mb_height * 16;
705
706     s->mb_num     = s->mb_width * s->mb_height;
707
708     s->block_wrap[0] =
709     s->block_wrap[1] =
710     s->block_wrap[2] =
711     s->block_wrap[3] = s->b8_stride;
712     s->block_wrap[4] =
713     s->block_wrap[5] = s->mb_stride;
714
715     y_size  = s->b8_stride * (2 * s->mb_height + 1);
716     c_size  = s->mb_stride * (s->mb_height + 1);
717     yc_size = y_size + 2   * c_size;
718
719     if (s->mb_height & 1)
720         yc_size += 2*s->b8_stride + 2*s->mb_stride;
721
722     FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_index2xy, (s->mb_num + 1) * sizeof(int),
723                       fail); // error resilience code looks cleaner with this
724     for (y = 0; y < s->mb_height; y++)
725         for (x = 0; x < s->mb_width; x++)
726             s->mb_index2xy[x + y * s->mb_width] = x + y * s->mb_stride;
727
728     s->mb_index2xy[s->mb_height * s->mb_width] = (s->mb_height - 1) * s->mb_stride + s->mb_width; // FIXME really needed?
729
730     if (s->encoding) {
731         /* Allocate MV tables */
732         FF_ALLOCZ_OR_GOTO(s->avctx, s->p_mv_table_base,                 mv_table_size * 2 * sizeof(int16_t), fail)
733         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_forw_mv_table_base,            mv_table_size * 2 * sizeof(int16_t), fail)
734         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_back_mv_table_base,            mv_table_size * 2 * sizeof(int16_t), fail)
735         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_forw_mv_table_base,      mv_table_size * 2 * sizeof(int16_t), fail)
736         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_back_mv_table_base,      mv_table_size * 2 * sizeof(int16_t), fail)
737         FF_ALLOCZ_OR_GOTO(s->avctx, s->b_direct_mv_table_base,          mv_table_size * 2 * sizeof(int16_t), fail)
738         s->p_mv_table            = s->p_mv_table_base + s->mb_stride + 1;
739         s->b_forw_mv_table       = s->b_forw_mv_table_base + s->mb_stride + 1;
740         s->b_back_mv_table       = s->b_back_mv_table_base + s->mb_stride + 1;
741         s->b_bidir_forw_mv_table = s->b_bidir_forw_mv_table_base + s->mb_stride + 1;
742         s->b_bidir_back_mv_table = s->b_bidir_back_mv_table_base + s->mb_stride + 1;
743         s->b_direct_mv_table     = s->b_direct_mv_table_base + s->mb_stride + 1;
744
745         /* Allocate MB type table */
746         FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_type, mb_array_size * sizeof(uint16_t), fail) // needed for encoding
747
748         FF_ALLOCZ_OR_GOTO(s->avctx, s->lambda_table, mb_array_size * sizeof(int), fail)
749
750         FF_ALLOC_OR_GOTO(s->avctx, s->cplx_tab,
751                          mb_array_size * sizeof(float), fail);
752         FF_ALLOC_OR_GOTO(s->avctx, s->bits_tab,
753                          mb_array_size * sizeof(float), fail);
754
755     }
756
757     if (s->codec_id == AV_CODEC_ID_MPEG4 ||
758         (s->avctx->flags & AV_CODEC_FLAG_INTERLACED_ME)) {
759         /* interlaced direct mode decoding tables */
760         for (i = 0; i < 2; i++) {
761             int j, k;
762             for (j = 0; j < 2; j++) {
763                 for (k = 0; k < 2; k++) {
764                     FF_ALLOCZ_OR_GOTO(s->avctx,
765                                       s->b_field_mv_table_base[i][j][k],
766                                       mv_table_size * 2 * sizeof(int16_t),
767                                       fail);
768                     s->b_field_mv_table[i][j][k] = s->b_field_mv_table_base[i][j][k] +
769                                                    s->mb_stride + 1;
770                 }
771                 FF_ALLOCZ_OR_GOTO(s->avctx, s->b_field_select_table [i][j], mb_array_size * 2 * sizeof(uint8_t), fail)
772                 FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_mv_table_base[i][j], mv_table_size * 2 * sizeof(int16_t), fail)
773                 s->p_field_mv_table[i][j] = s->p_field_mv_table_base[i][j] + s->mb_stride + 1;
774             }
775             FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_select_table[i], mb_array_size * 2 * sizeof(uint8_t), fail)
776         }
777     }
778     if (s->out_format == FMT_H263) {
779         /* cbp values */
780         FF_ALLOCZ_OR_GOTO(s->avctx, s->coded_block_base, y_size + (s->mb_height&1)*2*s->b8_stride, fail);
781         s->coded_block = s->coded_block_base + s->b8_stride + 1;
782
783         /* cbp, ac_pred, pred_dir */
784         FF_ALLOCZ_OR_GOTO(s->avctx, s->cbp_table     , mb_array_size * sizeof(uint8_t), fail);
785         FF_ALLOCZ_OR_GOTO(s->avctx, s->pred_dir_table, mb_array_size * sizeof(uint8_t), fail);
786     }
787
788     if (s->h263_pred || s->h263_plus || !s->encoding) {
789         /* dc values */
790         // MN: we need these for error resilience of intra-frames
791         FF_ALLOCZ_OR_GOTO(s->avctx, s->dc_val_base, yc_size * sizeof(int16_t), fail);
792         s->dc_val[0] = s->dc_val_base + s->b8_stride + 1;
793         s->dc_val[1] = s->dc_val_base + y_size + s->mb_stride + 1;
794         s->dc_val[2] = s->dc_val[1] + c_size;
795         for (i = 0; i < yc_size; i++)
796             s->dc_val_base[i] = 1024;
797     }
798
799     /* which mb is an intra block */
800     FF_ALLOCZ_OR_GOTO(s->avctx, s->mbintra_table, mb_array_size, fail);
801     memset(s->mbintra_table, 1, mb_array_size);
802
803     /* init macroblock skip table */
804     FF_ALLOCZ_OR_GOTO(s->avctx, s->mbskip_table, mb_array_size + 2, fail);
805     // Note the + 1 is for a quicker MPEG-4 slice_end detection
806
807     return ff_mpeg_er_init(s);
808 fail:
809     return AVERROR(ENOMEM);
810 }
811
812 static void clear_context(MpegEncContext *s)
813 {
814     int i, j, k;
815
816     memset(&s->next_picture, 0, sizeof(s->next_picture));
817     memset(&s->last_picture, 0, sizeof(s->last_picture));
818     memset(&s->current_picture, 0, sizeof(s->current_picture));
819     memset(&s->new_picture, 0, sizeof(s->new_picture));
820
821     memset(s->thread_context, 0, sizeof(s->thread_context));
822
823     s->me.map = NULL;
824     s->me.score_map = NULL;
825     s->dct_error_sum = NULL;
826     s->block = NULL;
827     s->blocks = NULL;
828     s->block32 = NULL;
829     memset(s->pblocks, 0, sizeof(s->pblocks));
830     s->dpcm_direction = 0;
831     s->dpcm_macroblock = NULL;
832     s->ac_val_base = NULL;
833     s->ac_val[0] =
834     s->ac_val[1] =
835     s->ac_val[2] =NULL;
836     s->sc.edge_emu_buffer = NULL;
837     s->me.scratchpad = NULL;
838     s->me.temp =
839     s->sc.rd_scratchpad =
840     s->sc.b_scratchpad =
841     s->sc.obmc_scratchpad = NULL;
842
843
844     s->bitstream_buffer = NULL;
845     s->allocated_bitstream_buffer_size = 0;
846     s->picture          = NULL;
847     s->mb_type          = NULL;
848     s->p_mv_table_base  = NULL;
849     s->b_forw_mv_table_base = NULL;
850     s->b_back_mv_table_base = NULL;
851     s->b_bidir_forw_mv_table_base = NULL;
852     s->b_bidir_back_mv_table_base = NULL;
853     s->b_direct_mv_table_base = NULL;
854     s->p_mv_table            = NULL;
855     s->b_forw_mv_table       = NULL;
856     s->b_back_mv_table       = NULL;
857     s->b_bidir_forw_mv_table = NULL;
858     s->b_bidir_back_mv_table = NULL;
859     s->b_direct_mv_table     = NULL;
860     for (i = 0; i < 2; i++) {
861         for (j = 0; j < 2; j++) {
862             for (k = 0; k < 2; k++) {
863                 s->b_field_mv_table_base[i][j][k] = NULL;
864                 s->b_field_mv_table[i][j][k] = NULL;
865             }
866             s->b_field_select_table[i][j] = NULL;
867             s->p_field_mv_table_base[i][j] = NULL;
868             s->p_field_mv_table[i][j] = NULL;
869         }
870         s->p_field_select_table[i] = NULL;
871     }
872
873     s->dc_val_base = NULL;
874     s->coded_block_base = NULL;
875     s->mbintra_table = NULL;
876     s->cbp_table = NULL;
877     s->pred_dir_table = NULL;
878
879     s->mbskip_table = NULL;
880
881     s->er.error_status_table = NULL;
882     s->er.er_temp_buffer = NULL;
883     s->mb_index2xy = NULL;
884     s->lambda_table = NULL;
885
886     s->cplx_tab = NULL;
887     s->bits_tab = NULL;
888 }
889
890 /**
891  * init common structure for both encoder and decoder.
892  * this assumes that some variables like width/height are already set
893  */
894 av_cold int ff_mpv_common_init(MpegEncContext *s)
895 {
896     int i, ret;
897     int nb_slices = (HAVE_THREADS &&
898                      s->avctx->active_thread_type & FF_THREAD_SLICE) ?
899                     s->avctx->thread_count : 1;
900
901     clear_context(s);
902
903     if (s->encoding && s->avctx->slices)
904         nb_slices = s->avctx->slices;
905
906     if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
907         s->mb_height = (s->height + 31) / 32 * 2;
908     else
909         s->mb_height = (s->height + 15) / 16;
910
911     if (s->avctx->pix_fmt == AV_PIX_FMT_NONE) {
912         av_log(s->avctx, AV_LOG_ERROR,
913                "decoding to AV_PIX_FMT_NONE is not supported.\n");
914         return -1;
915     }
916
917     if (nb_slices > MAX_THREADS || (nb_slices > s->mb_height && s->mb_height)) {
918         int max_slices;
919         if (s->mb_height)
920             max_slices = FFMIN(MAX_THREADS, s->mb_height);
921         else
922             max_slices = MAX_THREADS;
923         av_log(s->avctx, AV_LOG_WARNING, "too many threads/slices (%d),"
924                " reducing to %d\n", nb_slices, max_slices);
925         nb_slices = max_slices;
926     }
927
928     if ((s->width || s->height) &&
929         av_image_check_size(s->width, s->height, 0, s->avctx))
930         return -1;
931
932     dct_init(s);
933
934     /* set chroma shifts */
935     ret = av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
936                                            &s->chroma_x_shift,
937                                            &s->chroma_y_shift);
938     if (ret)
939         return ret;
940
941     FF_ALLOCZ_OR_GOTO(s->avctx, s->picture,
942                       MAX_PICTURE_COUNT * sizeof(Picture), fail);
943     for (i = 0; i < MAX_PICTURE_COUNT; i++) {
944         s->picture[i].f = av_frame_alloc();
945         if (!s->picture[i].f)
946             goto fail;
947     }
948     s->next_picture.f = av_frame_alloc();
949     if (!s->next_picture.f)
950         goto fail;
951     s->last_picture.f = av_frame_alloc();
952     if (!s->last_picture.f)
953         goto fail;
954     s->current_picture.f = av_frame_alloc();
955     if (!s->current_picture.f)
956         goto fail;
957     s->new_picture.f = av_frame_alloc();
958     if (!s->new_picture.f)
959         goto fail;
960
961     if (init_context_frame(s))
962         goto fail;
963
964     s->parse_context.state = -1;
965
966     s->context_initialized = 1;
967     memset(s->thread_context, 0, sizeof(s->thread_context));
968     s->thread_context[0]   = s;
969
970 //     if (s->width && s->height) {
971     if (nb_slices > 1) {
972         for (i = 0; i < nb_slices; i++) {
973             if (i) {
974                 s->thread_context[i] = av_memdup(s, sizeof(MpegEncContext));
975                 if (!s->thread_context[i])
976                     goto fail;
977             }
978             if (init_duplicate_context(s->thread_context[i]) < 0)
979                 goto fail;
980             s->thread_context[i]->start_mb_y =
981                 (s->mb_height * (i) + nb_slices / 2) / nb_slices;
982             s->thread_context[i]->end_mb_y   =
983                 (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices;
984         }
985     } else {
986         if (init_duplicate_context(s) < 0)
987             goto fail;
988         s->start_mb_y = 0;
989         s->end_mb_y   = s->mb_height;
990     }
991     s->slice_context_count = nb_slices;
992 //     }
993
994     return 0;
995  fail:
996     ff_mpv_common_end(s);
997     return -1;
998 }
999
1000 /**
1001  * Frees and resets MpegEncContext fields depending on the resolution.
1002  * Is used during resolution changes to avoid a full reinitialization of the
1003  * codec.
1004  */
1005 static void free_context_frame(MpegEncContext *s)
1006 {
1007     int i, j, k;
1008
1009     av_freep(&s->mb_type);
1010     av_freep(&s->p_mv_table_base);
1011     av_freep(&s->b_forw_mv_table_base);
1012     av_freep(&s->b_back_mv_table_base);
1013     av_freep(&s->b_bidir_forw_mv_table_base);
1014     av_freep(&s->b_bidir_back_mv_table_base);
1015     av_freep(&s->b_direct_mv_table_base);
1016     s->p_mv_table            = NULL;
1017     s->b_forw_mv_table       = NULL;
1018     s->b_back_mv_table       = NULL;
1019     s->b_bidir_forw_mv_table = NULL;
1020     s->b_bidir_back_mv_table = NULL;
1021     s->b_direct_mv_table     = NULL;
1022     for (i = 0; i < 2; i++) {
1023         for (j = 0; j < 2; j++) {
1024             for (k = 0; k < 2; k++) {
1025                 av_freep(&s->b_field_mv_table_base[i][j][k]);
1026                 s->b_field_mv_table[i][j][k] = NULL;
1027             }
1028             av_freep(&s->b_field_select_table[i][j]);
1029             av_freep(&s->p_field_mv_table_base[i][j]);
1030             s->p_field_mv_table[i][j] = NULL;
1031         }
1032         av_freep(&s->p_field_select_table[i]);
1033     }
1034
1035     av_freep(&s->dc_val_base);
1036     av_freep(&s->coded_block_base);
1037     av_freep(&s->mbintra_table);
1038     av_freep(&s->cbp_table);
1039     av_freep(&s->pred_dir_table);
1040
1041     av_freep(&s->mbskip_table);
1042
1043     av_freep(&s->er.error_status_table);
1044     av_freep(&s->er.er_temp_buffer);
1045     av_freep(&s->mb_index2xy);
1046     av_freep(&s->lambda_table);
1047
1048     av_freep(&s->cplx_tab);
1049     av_freep(&s->bits_tab);
1050
1051     s->linesize = s->uvlinesize = 0;
1052 }
1053
1054 int ff_mpv_common_frame_size_change(MpegEncContext *s)
1055 {
1056     int i, err = 0;
1057
1058     if (!s->context_initialized)
1059         return AVERROR(EINVAL);
1060
1061     if (s->slice_context_count > 1) {
1062         for (i = 0; i < s->slice_context_count; i++) {
1063             free_duplicate_context(s->thread_context[i]);
1064         }
1065         for (i = 1; i < s->slice_context_count; i++) {
1066             av_freep(&s->thread_context[i]);
1067         }
1068     } else
1069         free_duplicate_context(s);
1070
1071     free_context_frame(s);
1072
1073     if (s->picture)
1074         for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1075                 s->picture[i].needs_realloc = 1;
1076         }
1077
1078     s->last_picture_ptr         =
1079     s->next_picture_ptr         =
1080     s->current_picture_ptr      = NULL;
1081
1082     // init
1083     if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO && !s->progressive_sequence)
1084         s->mb_height = (s->height + 31) / 32 * 2;
1085     else
1086         s->mb_height = (s->height + 15) / 16;
1087
1088     if ((s->width || s->height) &&
1089         (err = av_image_check_size(s->width, s->height, 0, s->avctx)) < 0)
1090         goto fail;
1091
1092     if ((err = init_context_frame(s)))
1093         goto fail;
1094
1095     memset(s->thread_context, 0, sizeof(s->thread_context));
1096     s->thread_context[0]   = s;
1097
1098     if (s->width && s->height) {
1099         int nb_slices = s->slice_context_count;
1100         if (nb_slices > 1) {
1101             for (i = 0; i < nb_slices; i++) {
1102                 if (i) {
1103                     s->thread_context[i] = av_memdup(s, sizeof(MpegEncContext));
1104                     if (!s->thread_context[i]) {
1105                         err = AVERROR(ENOMEM);
1106                         goto fail;
1107                     }
1108                 }
1109                 if ((err = init_duplicate_context(s->thread_context[i])) < 0)
1110                     goto fail;
1111                 s->thread_context[i]->start_mb_y =
1112                     (s->mb_height * (i) + nb_slices / 2) / nb_slices;
1113                 s->thread_context[i]->end_mb_y   =
1114                     (s->mb_height * (i + 1) + nb_slices / 2) / nb_slices;
1115             }
1116         } else {
1117             err = init_duplicate_context(s);
1118             if (err < 0)
1119                 goto fail;
1120             s->start_mb_y = 0;
1121             s->end_mb_y   = s->mb_height;
1122         }
1123         s->slice_context_count = nb_slices;
1124     }
1125
1126     return 0;
1127  fail:
1128     ff_mpv_common_end(s);
1129     return err;
1130 }
1131
1132 /* init common structure for both encoder and decoder */
1133 void ff_mpv_common_end(MpegEncContext *s)
1134 {
1135     int i;
1136
1137     if (!s)
1138         return ;
1139
1140     if (s->slice_context_count > 1) {
1141         for (i = 0; i < s->slice_context_count; i++) {
1142             free_duplicate_context(s->thread_context[i]);
1143         }
1144         for (i = 1; i < s->slice_context_count; i++) {
1145             av_freep(&s->thread_context[i]);
1146         }
1147         s->slice_context_count = 1;
1148     } else free_duplicate_context(s);
1149
1150     av_freep(&s->parse_context.buffer);
1151     s->parse_context.buffer_size = 0;
1152
1153     av_freep(&s->bitstream_buffer);
1154     s->allocated_bitstream_buffer_size = 0;
1155
1156     if (s->picture) {
1157         for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1158             ff_free_picture_tables(&s->picture[i]);
1159             ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
1160             av_frame_free(&s->picture[i].f);
1161         }
1162     }
1163     av_freep(&s->picture);
1164     ff_free_picture_tables(&s->last_picture);
1165     ff_mpeg_unref_picture(s->avctx, &s->last_picture);
1166     av_frame_free(&s->last_picture.f);
1167     ff_free_picture_tables(&s->current_picture);
1168     ff_mpeg_unref_picture(s->avctx, &s->current_picture);
1169     av_frame_free(&s->current_picture.f);
1170     ff_free_picture_tables(&s->next_picture);
1171     ff_mpeg_unref_picture(s->avctx, &s->next_picture);
1172     av_frame_free(&s->next_picture.f);
1173     ff_free_picture_tables(&s->new_picture);
1174     ff_mpeg_unref_picture(s->avctx, &s->new_picture);
1175     av_frame_free(&s->new_picture.f);
1176
1177     free_context_frame(s);
1178
1179     s->context_initialized      = 0;
1180     s->last_picture_ptr         =
1181     s->next_picture_ptr         =
1182     s->current_picture_ptr      = NULL;
1183     s->linesize = s->uvlinesize = 0;
1184 }
1185
1186
1187 static void gray_frame(AVFrame *frame)
1188 {
1189     int i, h_chroma_shift, v_chroma_shift;
1190
1191     av_pix_fmt_get_chroma_sub_sample(frame->format, &h_chroma_shift, &v_chroma_shift);
1192
1193     for(i=0; i<frame->height; i++)
1194         memset(frame->data[0] + frame->linesize[0]*i, 0x80, frame->width);
1195     for(i=0; i<AV_CEIL_RSHIFT(frame->height, v_chroma_shift); i++) {
1196         memset(frame->data[1] + frame->linesize[1]*i,
1197                0x80, AV_CEIL_RSHIFT(frame->width, h_chroma_shift));
1198         memset(frame->data[2] + frame->linesize[2]*i,
1199                0x80, AV_CEIL_RSHIFT(frame->width, h_chroma_shift));
1200     }
1201 }
1202
1203 /**
1204  * generic function called after decoding
1205  * the header and before a frame is decoded.
1206  */
1207 int ff_mpv_frame_start(MpegEncContext *s, AVCodecContext *avctx)
1208 {
1209     int i, ret;
1210     Picture *pic;
1211     s->mb_skipped = 0;
1212
1213     if (!ff_thread_can_start_frame(avctx)) {
1214         av_log(avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
1215         return -1;
1216     }
1217
1218     /* mark & release old frames */
1219     if (s->pict_type != AV_PICTURE_TYPE_B && s->last_picture_ptr &&
1220         s->last_picture_ptr != s->next_picture_ptr &&
1221         s->last_picture_ptr->f->buf[0]) {
1222         ff_mpeg_unref_picture(s->avctx, s->last_picture_ptr);
1223     }
1224
1225     /* release forgotten pictures */
1226     /* if (MPEG-124 / H.263) */
1227     for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1228         if (&s->picture[i] != s->last_picture_ptr &&
1229             &s->picture[i] != s->next_picture_ptr &&
1230             s->picture[i].reference && !s->picture[i].needs_realloc) {
1231             ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
1232         }
1233     }
1234
1235     ff_mpeg_unref_picture(s->avctx, &s->current_picture);
1236     ff_mpeg_unref_picture(s->avctx, &s->last_picture);
1237     ff_mpeg_unref_picture(s->avctx, &s->next_picture);
1238
1239     /* release non reference frames */
1240     for (i = 0; i < MAX_PICTURE_COUNT; i++) {
1241         if (!s->picture[i].reference)
1242             ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
1243     }
1244
1245     if (s->current_picture_ptr && !s->current_picture_ptr->f->buf[0]) {
1246         // we already have an unused image
1247         // (maybe it was set before reading the header)
1248         pic = s->current_picture_ptr;
1249     } else {
1250         i   = ff_find_unused_picture(s->avctx, s->picture, 0);
1251         if (i < 0) {
1252             av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1253             return i;
1254         }
1255         pic = &s->picture[i];
1256     }
1257
1258     pic->reference = 0;
1259     if (!s->droppable) {
1260         if (s->pict_type != AV_PICTURE_TYPE_B)
1261             pic->reference = 3;
1262     }
1263
1264     pic->f->coded_picture_number = s->coded_picture_number++;
1265
1266     if (alloc_picture(s, pic, 0) < 0)
1267         return -1;
1268
1269     s->current_picture_ptr = pic;
1270     // FIXME use only the vars from current_pic
1271     s->current_picture_ptr->f->top_field_first = s->top_field_first;
1272     if (s->codec_id == AV_CODEC_ID_MPEG1VIDEO ||
1273         s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
1274         if (s->picture_structure != PICT_FRAME)
1275             s->current_picture_ptr->f->top_field_first =
1276                 (s->picture_structure == PICT_TOP_FIELD) == s->first_field;
1277     }
1278     s->current_picture_ptr->f->interlaced_frame = !s->progressive_frame &&
1279                                                  !s->progressive_sequence;
1280     s->current_picture_ptr->field_picture      =  s->picture_structure != PICT_FRAME;
1281
1282     s->current_picture_ptr->f->pict_type = s->pict_type;
1283     // if (s->avctx->flags && AV_CODEC_FLAG_QSCALE)
1284     //     s->current_picture_ptr->quality = s->new_picture_ptr->quality;
1285     s->current_picture_ptr->f->key_frame = s->pict_type == AV_PICTURE_TYPE_I;
1286
1287     if ((ret = ff_mpeg_ref_picture(s->avctx, &s->current_picture,
1288                                    s->current_picture_ptr)) < 0)
1289         return ret;
1290
1291     if (s->pict_type != AV_PICTURE_TYPE_B) {
1292         s->last_picture_ptr = s->next_picture_ptr;
1293         if (!s->droppable)
1294             s->next_picture_ptr = s->current_picture_ptr;
1295     }
1296     ff_dlog(s->avctx, "L%p N%p C%p L%p N%p C%p type:%d drop:%d\n",
1297             s->last_picture_ptr, s->next_picture_ptr,s->current_picture_ptr,
1298             s->last_picture_ptr    ? s->last_picture_ptr->f->data[0]    : NULL,
1299             s->next_picture_ptr    ? s->next_picture_ptr->f->data[0]    : NULL,
1300             s->current_picture_ptr ? s->current_picture_ptr->f->data[0] : NULL,
1301             s->pict_type, s->droppable);
1302
1303     if ((!s->last_picture_ptr || !s->last_picture_ptr->f->buf[0]) &&
1304         (s->pict_type != AV_PICTURE_TYPE_I)) {
1305         int h_chroma_shift, v_chroma_shift;
1306         av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
1307                                          &h_chroma_shift, &v_chroma_shift);
1308         if (s->pict_type == AV_PICTURE_TYPE_B && s->next_picture_ptr && s->next_picture_ptr->f->buf[0])
1309             av_log(avctx, AV_LOG_DEBUG,
1310                    "allocating dummy last picture for B frame\n");
1311         else if (s->pict_type != AV_PICTURE_TYPE_I)
1312             av_log(avctx, AV_LOG_ERROR,
1313                    "warning: first frame is no keyframe\n");
1314
1315         /* Allocate a dummy frame */
1316         i = ff_find_unused_picture(s->avctx, s->picture, 0);
1317         if (i < 0) {
1318             av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1319             return i;
1320         }
1321         s->last_picture_ptr = &s->picture[i];
1322
1323         s->last_picture_ptr->reference   = 3;
1324         s->last_picture_ptr->f->key_frame = 0;
1325         s->last_picture_ptr->f->pict_type = AV_PICTURE_TYPE_P;
1326
1327         if (alloc_picture(s, s->last_picture_ptr, 0) < 0) {
1328             s->last_picture_ptr = NULL;
1329             return -1;
1330         }
1331
1332         if (!avctx->hwaccel) {
1333             for(i=0; i<avctx->height; i++)
1334                 memset(s->last_picture_ptr->f->data[0] + s->last_picture_ptr->f->linesize[0]*i,
1335                        0x80, avctx->width);
1336             if (s->last_picture_ptr->f->data[2]) {
1337                 for(i=0; i<AV_CEIL_RSHIFT(avctx->height, v_chroma_shift); i++) {
1338                     memset(s->last_picture_ptr->f->data[1] + s->last_picture_ptr->f->linesize[1]*i,
1339                         0x80, AV_CEIL_RSHIFT(avctx->width, h_chroma_shift));
1340                     memset(s->last_picture_ptr->f->data[2] + s->last_picture_ptr->f->linesize[2]*i,
1341                         0x80, AV_CEIL_RSHIFT(avctx->width, h_chroma_shift));
1342                 }
1343             }
1344
1345             if(s->codec_id == AV_CODEC_ID_FLV1 || s->codec_id == AV_CODEC_ID_H263){
1346                 for(i=0; i<avctx->height; i++)
1347                 memset(s->last_picture_ptr->f->data[0] + s->last_picture_ptr->f->linesize[0]*i, 16, avctx->width);
1348             }
1349         }
1350
1351         ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 0);
1352         ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 1);
1353     }
1354     if ((!s->next_picture_ptr || !s->next_picture_ptr->f->buf[0]) &&
1355         s->pict_type == AV_PICTURE_TYPE_B) {
1356         /* Allocate a dummy frame */
1357         i = ff_find_unused_picture(s->avctx, s->picture, 0);
1358         if (i < 0) {
1359             av_log(s->avctx, AV_LOG_ERROR, "no frame buffer available\n");
1360             return i;
1361         }
1362         s->next_picture_ptr = &s->picture[i];
1363
1364         s->next_picture_ptr->reference   = 3;
1365         s->next_picture_ptr->f->key_frame = 0;
1366         s->next_picture_ptr->f->pict_type = AV_PICTURE_TYPE_P;
1367
1368         if (alloc_picture(s, s->next_picture_ptr, 0) < 0) {
1369             s->next_picture_ptr = NULL;
1370             return -1;
1371         }
1372         ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 0);
1373         ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 1);
1374     }
1375
1376 #if 0 // BUFREF-FIXME
1377     memset(s->last_picture.f->data, 0, sizeof(s->last_picture.f->data));
1378     memset(s->next_picture.f->data, 0, sizeof(s->next_picture.f->data));
1379 #endif
1380     if (s->last_picture_ptr) {
1381         if (s->last_picture_ptr->f->buf[0] &&
1382             (ret = ff_mpeg_ref_picture(s->avctx, &s->last_picture,
1383                                        s->last_picture_ptr)) < 0)
1384             return ret;
1385     }
1386     if (s->next_picture_ptr) {
1387         if (s->next_picture_ptr->f->buf[0] &&
1388             (ret = ff_mpeg_ref_picture(s->avctx, &s->next_picture,
1389                                        s->next_picture_ptr)) < 0)
1390             return ret;
1391     }
1392
1393     av_assert0(s->pict_type == AV_PICTURE_TYPE_I || (s->last_picture_ptr &&
1394                                                  s->last_picture_ptr->f->buf[0]));
1395
1396     if (s->picture_structure!= PICT_FRAME) {
1397         int i;
1398         for (i = 0; i < 4; i++) {
1399             if (s->picture_structure == PICT_BOTTOM_FIELD) {
1400                 s->current_picture.f->data[i] +=
1401                     s->current_picture.f->linesize[i];
1402             }
1403             s->current_picture.f->linesize[i] *= 2;
1404             s->last_picture.f->linesize[i]    *= 2;
1405             s->next_picture.f->linesize[i]    *= 2;
1406         }
1407     }
1408
1409     /* set dequantizer, we can't do it during init as
1410      * it might change for MPEG-4 and we can't do it in the header
1411      * decode as init is not called for MPEG-4 there yet */
1412     if (s->mpeg_quant || s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
1413         s->dct_unquantize_intra = s->dct_unquantize_mpeg2_intra;
1414         s->dct_unquantize_inter = s->dct_unquantize_mpeg2_inter;
1415     } else if (s->out_format == FMT_H263 || s->out_format == FMT_H261) {
1416         s->dct_unquantize_intra = s->dct_unquantize_h263_intra;
1417         s->dct_unquantize_inter = s->dct_unquantize_h263_inter;
1418     } else {
1419         s->dct_unquantize_intra = s->dct_unquantize_mpeg1_intra;
1420         s->dct_unquantize_inter = s->dct_unquantize_mpeg1_inter;
1421     }
1422
1423     if (s->avctx->debug & FF_DEBUG_NOMC) {
1424         gray_frame(s->current_picture_ptr->f);
1425     }
1426
1427     return 0;
1428 }
1429
1430 /* called after a frame has been decoded. */
1431 void ff_mpv_frame_end(MpegEncContext *s)
1432 {
1433     emms_c();
1434
1435     if (s->current_picture.reference)
1436         ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
1437 }
1438
1439 void ff_print_debug_info(MpegEncContext *s, Picture *p, AVFrame *pict)
1440 {
1441     ff_print_debug_info2(s->avctx, pict, s->mbskip_table, p->mb_type,
1442                          p->qscale_table, p->motion_val, &s->low_delay,
1443                          s->mb_width, s->mb_height, s->mb_stride, s->quarter_sample);
1444 }
1445
1446 int ff_mpv_export_qp_table(MpegEncContext *s, AVFrame *f, Picture *p, int qp_type)
1447 {
1448     AVBufferRef *ref = av_buffer_ref(p->qscale_table_buf);
1449     int offset = 2*s->mb_stride + 1;
1450     if(!ref)
1451         return AVERROR(ENOMEM);
1452     av_assert0(ref->size >= offset + s->mb_stride * ((f->height+15)/16));
1453     ref->size -= offset;
1454     ref->data += offset;
1455     return av_frame_set_qp_table(f, ref, s->mb_stride, qp_type);
1456 }
1457
1458 static inline int hpel_motion_lowres(MpegEncContext *s,
1459                                      uint8_t *dest, uint8_t *src,
1460                                      int field_based, int field_select,
1461                                      int src_x, int src_y,
1462                                      int width, int height, ptrdiff_t stride,
1463                                      int h_edge_pos, int v_edge_pos,
1464                                      int w, int h, h264_chroma_mc_func *pix_op,
1465                                      int motion_x, int motion_y)
1466 {
1467     const int lowres   = s->avctx->lowres;
1468     const int op_index = FFMIN(lowres, 3);
1469     const int s_mask   = (2 << lowres) - 1;
1470     int emu = 0;
1471     int sx, sy;
1472
1473     if (s->quarter_sample) {
1474         motion_x /= 2;
1475         motion_y /= 2;
1476     }
1477
1478     sx = motion_x & s_mask;
1479     sy = motion_y & s_mask;
1480     src_x += motion_x >> lowres + 1;
1481     src_y += motion_y >> lowres + 1;
1482
1483     src   += src_y * stride + src_x;
1484
1485     if ((unsigned)src_x > FFMAX( h_edge_pos - (!!sx) - w,                 0) ||
1486         (unsigned)src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
1487         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, src,
1488                                  s->linesize, s->linesize,
1489                                  w + 1, (h + 1) << field_based,
1490                                  src_x, src_y   << field_based,
1491                                  h_edge_pos, v_edge_pos);
1492         src = s->sc.edge_emu_buffer;
1493         emu = 1;
1494     }
1495
1496     sx = (sx << 2) >> lowres;
1497     sy = (sy << 2) >> lowres;
1498     if (field_select)
1499         src += s->linesize;
1500     pix_op[op_index](dest, src, stride, h, sx, sy);
1501     return emu;
1502 }
1503
1504 /* apply one mpeg motion vector to the three components */
1505 static av_always_inline void mpeg_motion_lowres(MpegEncContext *s,
1506                                                 uint8_t *dest_y,
1507                                                 uint8_t *dest_cb,
1508                                                 uint8_t *dest_cr,
1509                                                 int field_based,
1510                                                 int bottom_field,
1511                                                 int field_select,
1512                                                 uint8_t **ref_picture,
1513                                                 h264_chroma_mc_func *pix_op,
1514                                                 int motion_x, int motion_y,
1515                                                 int h, int mb_y)
1516 {
1517     uint8_t *ptr_y, *ptr_cb, *ptr_cr;
1518     int mx, my, src_x, src_y, uvsrc_x, uvsrc_y, sx, sy, uvsx, uvsy;
1519     ptrdiff_t uvlinesize, linesize;
1520     const int lowres     = s->avctx->lowres;
1521     const int op_index   = FFMIN(lowres-1+s->chroma_x_shift, 3);
1522     const int block_s    = 8>>lowres;
1523     const int s_mask     = (2 << lowres) - 1;
1524     const int h_edge_pos = s->h_edge_pos >> lowres;
1525     const int v_edge_pos = s->v_edge_pos >> lowres;
1526     linesize   = s->current_picture.f->linesize[0] << field_based;
1527     uvlinesize = s->current_picture.f->linesize[1] << field_based;
1528
1529     // FIXME obviously not perfect but qpel will not work in lowres anyway
1530     if (s->quarter_sample) {
1531         motion_x /= 2;
1532         motion_y /= 2;
1533     }
1534
1535     if(field_based){
1536         motion_y += (bottom_field - field_select)*((1 << lowres)-1);
1537     }
1538
1539     sx = motion_x & s_mask;
1540     sy = motion_y & s_mask;
1541     src_x = s->mb_x * 2 * block_s + (motion_x >> lowres + 1);
1542     src_y = (mb_y * 2 * block_s >> field_based) + (motion_y >> lowres + 1);
1543
1544     if (s->out_format == FMT_H263) {
1545         uvsx    = ((motion_x >> 1) & s_mask) | (sx & 1);
1546         uvsy    = ((motion_y >> 1) & s_mask) | (sy & 1);
1547         uvsrc_x = src_x >> 1;
1548         uvsrc_y = src_y >> 1;
1549     } else if (s->out_format == FMT_H261) {
1550         // even chroma mv's are full pel in H261
1551         mx      = motion_x / 4;
1552         my      = motion_y / 4;
1553         uvsx    = (2 * mx) & s_mask;
1554         uvsy    = (2 * my) & s_mask;
1555         uvsrc_x = s->mb_x * block_s + (mx >> lowres);
1556         uvsrc_y =    mb_y * block_s + (my >> lowres);
1557     } else {
1558         if(s->chroma_y_shift){
1559             mx      = motion_x / 2;
1560             my      = motion_y / 2;
1561             uvsx    = mx & s_mask;
1562             uvsy    = my & s_mask;
1563             uvsrc_x = s->mb_x * block_s                 + (mx >> lowres + 1);
1564             uvsrc_y =   (mb_y * block_s >> field_based) + (my >> lowres + 1);
1565         } else {
1566             if(s->chroma_x_shift){
1567             //Chroma422
1568                 mx = motion_x / 2;
1569                 uvsx = mx & s_mask;
1570                 uvsy = motion_y & s_mask;
1571                 uvsrc_y = src_y;
1572                 uvsrc_x = s->mb_x*block_s               + (mx >> (lowres+1));
1573             } else {
1574             //Chroma444
1575                 uvsx = motion_x & s_mask;
1576                 uvsy = motion_y & s_mask;
1577                 uvsrc_x = src_x;
1578                 uvsrc_y = src_y;
1579             }
1580         }
1581     }
1582
1583     ptr_y  = ref_picture[0] + src_y   * linesize   + src_x;
1584     ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
1585     ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
1586
1587     if ((unsigned) src_x > FFMAX( h_edge_pos - (!!sx) - 2 * block_s,       0) || uvsrc_y<0 ||
1588         (unsigned) src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
1589         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr_y,
1590                                  linesize >> field_based, linesize >> field_based,
1591                                  17, 17 + field_based,
1592                                 src_x, src_y << field_based, h_edge_pos,
1593                                 v_edge_pos);
1594         ptr_y = s->sc.edge_emu_buffer;
1595         if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
1596             uint8_t *ubuf = s->sc.edge_emu_buffer + 18 * s->linesize;
1597             uint8_t *vbuf =ubuf + 10 * s->uvlinesize;
1598             if (s->workaround_bugs & FF_BUG_IEDGE)
1599                 vbuf -= s->uvlinesize;
1600             s->vdsp.emulated_edge_mc(ubuf,  ptr_cb,
1601                                      uvlinesize >> field_based, uvlinesize >> field_based,
1602                                      9, 9 + field_based,
1603                                     uvsrc_x, uvsrc_y << field_based,
1604                                     h_edge_pos >> 1, v_edge_pos >> 1);
1605             s->vdsp.emulated_edge_mc(vbuf,  ptr_cr,
1606                                      uvlinesize >> field_based,uvlinesize >> field_based,
1607                                      9, 9 + field_based,
1608                                     uvsrc_x, uvsrc_y << field_based,
1609                                     h_edge_pos >> 1, v_edge_pos >> 1);
1610             ptr_cb = ubuf;
1611             ptr_cr = vbuf;
1612         }
1613     }
1614
1615     // FIXME use this for field pix too instead of the obnoxious hack which changes picture.f->data
1616     if (bottom_field) {
1617         dest_y  += s->linesize;
1618         dest_cb += s->uvlinesize;
1619         dest_cr += s->uvlinesize;
1620     }
1621
1622     if (field_select) {
1623         ptr_y   += s->linesize;
1624         ptr_cb  += s->uvlinesize;
1625         ptr_cr  += s->uvlinesize;
1626     }
1627
1628     sx = (sx << 2) >> lowres;
1629     sy = (sy << 2) >> lowres;
1630     pix_op[lowres - 1](dest_y, ptr_y, linesize, h, sx, sy);
1631
1632     if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
1633         int hc = s->chroma_y_shift ? (h+1-bottom_field)>>1 : h;
1634         uvsx = (uvsx << 2) >> lowres;
1635         uvsy = (uvsy << 2) >> lowres;
1636         if (hc) {
1637             pix_op[op_index](dest_cb, ptr_cb, uvlinesize, hc, uvsx, uvsy);
1638             pix_op[op_index](dest_cr, ptr_cr, uvlinesize, hc, uvsx, uvsy);
1639         }
1640     }
1641     // FIXME h261 lowres loop filter
1642 }
1643
1644 static inline void chroma_4mv_motion_lowres(MpegEncContext *s,
1645                                             uint8_t *dest_cb, uint8_t *dest_cr,
1646                                             uint8_t **ref_picture,
1647                                             h264_chroma_mc_func * pix_op,
1648                                             int mx, int my)
1649 {
1650     const int lowres     = s->avctx->lowres;
1651     const int op_index   = FFMIN(lowres, 3);
1652     const int block_s    = 8 >> lowres;
1653     const int s_mask     = (2 << lowres) - 1;
1654     const int h_edge_pos = s->h_edge_pos >> lowres + 1;
1655     const int v_edge_pos = s->v_edge_pos >> lowres + 1;
1656     int emu = 0, src_x, src_y, sx, sy;
1657     ptrdiff_t offset;
1658     uint8_t *ptr;
1659
1660     if (s->quarter_sample) {
1661         mx /= 2;
1662         my /= 2;
1663     }
1664
1665     /* In case of 8X8, we construct a single chroma motion vector
1666        with a special rounding */
1667     mx = ff_h263_round_chroma(mx);
1668     my = ff_h263_round_chroma(my);
1669
1670     sx = mx & s_mask;
1671     sy = my & s_mask;
1672     src_x = s->mb_x * block_s + (mx >> lowres + 1);
1673     src_y = s->mb_y * block_s + (my >> lowres + 1);
1674
1675     offset = src_y * s->uvlinesize + src_x;
1676     ptr = ref_picture[1] + offset;
1677     if ((unsigned) src_x > FFMAX(h_edge_pos - (!!sx) - block_s, 0) ||
1678         (unsigned) src_y > FFMAX(v_edge_pos - (!!sy) - block_s, 0)) {
1679         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
1680                                  s->uvlinesize, s->uvlinesize,
1681                                  9, 9,
1682                                  src_x, src_y, h_edge_pos, v_edge_pos);
1683         ptr = s->sc.edge_emu_buffer;
1684         emu = 1;
1685     }
1686     sx = (sx << 2) >> lowres;
1687     sy = (sy << 2) >> lowres;
1688     pix_op[op_index](dest_cb, ptr, s->uvlinesize, block_s, sx, sy);
1689
1690     ptr = ref_picture[2] + offset;
1691     if (emu) {
1692         s->vdsp.emulated_edge_mc(s->sc.edge_emu_buffer, ptr,
1693                                  s->uvlinesize, s->uvlinesize,
1694                                  9, 9,
1695                                  src_x, src_y, h_edge_pos, v_edge_pos);
1696         ptr = s->sc.edge_emu_buffer;
1697     }
1698     pix_op[op_index](dest_cr, ptr, s->uvlinesize, block_s, sx, sy);
1699 }
1700
1701 /**
1702  * motion compensation of a single macroblock
1703  * @param s context
1704  * @param dest_y luma destination pointer
1705  * @param dest_cb chroma cb/u destination pointer
1706  * @param dest_cr chroma cr/v destination pointer
1707  * @param dir direction (0->forward, 1->backward)
1708  * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
1709  * @param pix_op halfpel motion compensation function (average or put normally)
1710  * the motion vectors are taken from s->mv and the MV type from s->mv_type
1711  */
1712 static inline void MPV_motion_lowres(MpegEncContext *s,
1713                                      uint8_t *dest_y, uint8_t *dest_cb,
1714                                      uint8_t *dest_cr,
1715                                      int dir, uint8_t **ref_picture,
1716                                      h264_chroma_mc_func *pix_op)
1717 {
1718     int mx, my;
1719     int mb_x, mb_y, i;
1720     const int lowres  = s->avctx->lowres;
1721     const int block_s = 8 >>lowres;
1722
1723     mb_x = s->mb_x;
1724     mb_y = s->mb_y;
1725
1726     switch (s->mv_type) {
1727     case MV_TYPE_16X16:
1728         mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1729                            0, 0, 0,
1730                            ref_picture, pix_op,
1731                            s->mv[dir][0][0], s->mv[dir][0][1],
1732                            2 * block_s, mb_y);
1733         break;
1734     case MV_TYPE_8X8:
1735         mx = 0;
1736         my = 0;
1737         for (i = 0; i < 4; i++) {
1738             hpel_motion_lowres(s, dest_y + ((i & 1) + (i >> 1) *
1739                                s->linesize) * block_s,
1740                                ref_picture[0], 0, 0,
1741                                (2 * mb_x + (i & 1)) * block_s,
1742                                (2 * mb_y + (i >> 1)) * block_s,
1743                                s->width, s->height, s->linesize,
1744                                s->h_edge_pos >> lowres, s->v_edge_pos >> lowres,
1745                                block_s, block_s, pix_op,
1746                                s->mv[dir][i][0], s->mv[dir][i][1]);
1747
1748             mx += s->mv[dir][i][0];
1749             my += s->mv[dir][i][1];
1750         }
1751
1752         if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY))
1753             chroma_4mv_motion_lowres(s, dest_cb, dest_cr, ref_picture,
1754                                      pix_op, mx, my);
1755         break;
1756     case MV_TYPE_FIELD:
1757         if (s->picture_structure == PICT_FRAME) {
1758             /* top field */
1759             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1760                                1, 0, s->field_select[dir][0],
1761                                ref_picture, pix_op,
1762                                s->mv[dir][0][0], s->mv[dir][0][1],
1763                                block_s, mb_y);
1764             /* bottom field */
1765             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1766                                1, 1, s->field_select[dir][1],
1767                                ref_picture, pix_op,
1768                                s->mv[dir][1][0], s->mv[dir][1][1],
1769                                block_s, mb_y);
1770         } else {
1771             if (s->picture_structure != s->field_select[dir][0] + 1 &&
1772                 s->pict_type != AV_PICTURE_TYPE_B && !s->first_field) {
1773                 ref_picture = s->current_picture_ptr->f->data;
1774
1775             }
1776             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1777                                0, 0, s->field_select[dir][0],
1778                                ref_picture, pix_op,
1779                                s->mv[dir][0][0],
1780                                s->mv[dir][0][1], 2 * block_s, mb_y >> 1);
1781             }
1782         break;
1783     case MV_TYPE_16X8:
1784         for (i = 0; i < 2; i++) {
1785             uint8_t **ref2picture;
1786
1787             if (s->picture_structure == s->field_select[dir][i] + 1 ||
1788                 s->pict_type == AV_PICTURE_TYPE_B || s->first_field) {
1789                 ref2picture = ref_picture;
1790             } else {
1791                 ref2picture = s->current_picture_ptr->f->data;
1792             }
1793
1794             mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1795                                0, 0, s->field_select[dir][i],
1796                                ref2picture, pix_op,
1797                                s->mv[dir][i][0], s->mv[dir][i][1] +
1798                                2 * block_s * i, block_s, mb_y >> 1);
1799
1800             dest_y  +=  2 * block_s *  s->linesize;
1801             dest_cb += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize;
1802             dest_cr += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize;
1803         }
1804         break;
1805     case MV_TYPE_DMV:
1806         if (s->picture_structure == PICT_FRAME) {
1807             for (i = 0; i < 2; i++) {
1808                 int j;
1809                 for (j = 0; j < 2; j++) {
1810                     mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1811                                        1, j, j ^ i,
1812                                        ref_picture, pix_op,
1813                                        s->mv[dir][2 * i + j][0],
1814                                        s->mv[dir][2 * i + j][1],
1815                                        block_s, mb_y);
1816                 }
1817                 pix_op = s->h264chroma.avg_h264_chroma_pixels_tab;
1818             }
1819         } else {
1820             for (i = 0; i < 2; i++) {
1821                 mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
1822                                    0, 0, s->picture_structure != i + 1,
1823                                    ref_picture, pix_op,
1824                                    s->mv[dir][2 * i][0],s->mv[dir][2 * i][1],
1825                                    2 * block_s, mb_y >> 1);
1826
1827                 // after put we make avg of the same block
1828                 pix_op = s->h264chroma.avg_h264_chroma_pixels_tab;
1829
1830                 // opposite parity is always in the same
1831                 // frame if this is second field
1832                 if (!s->first_field) {
1833                     ref_picture = s->current_picture_ptr->f->data;
1834                 }
1835             }
1836         }
1837         break;
1838     default:
1839         av_assert2(0);
1840     }
1841 }
1842
1843 /**
1844  * find the lowest MB row referenced in the MVs
1845  */
1846 static int lowest_referenced_row(MpegEncContext *s, int dir)
1847 {
1848     int my_max = INT_MIN, my_min = INT_MAX, qpel_shift = !s->quarter_sample;
1849     int my, off, i, mvs;
1850
1851     if (s->picture_structure != PICT_FRAME || s->mcsel)
1852         goto unhandled;
1853
1854     switch (s->mv_type) {
1855         case MV_TYPE_16X16:
1856             mvs = 1;
1857             break;
1858         case MV_TYPE_16X8:
1859             mvs = 2;
1860             break;
1861         case MV_TYPE_8X8:
1862             mvs = 4;
1863             break;
1864         default:
1865             goto unhandled;
1866     }
1867
1868     for (i = 0; i < mvs; i++) {
1869         my = s->mv[dir][i][1];
1870         my_max = FFMAX(my_max, my);
1871         my_min = FFMIN(my_min, my);
1872     }
1873
1874     off = ((FFMAX(-my_min, my_max)<<qpel_shift) + 63) >> 6;
1875
1876     return av_clip(s->mb_y + off, 0, s->mb_height - 1);
1877 unhandled:
1878     return s->mb_height-1;
1879 }
1880
1881 /* put block[] to dest[] */
1882 static inline void put_dct(MpegEncContext *s,
1883                            int16_t *block, int i, uint8_t *dest, int line_size, int qscale)
1884 {
1885     s->dct_unquantize_intra(s, block, i, qscale);
1886     s->idsp.idct_put(dest, line_size, block);
1887 }
1888
1889 /* add block[] to dest[] */
1890 static inline void add_dct(MpegEncContext *s,
1891                            int16_t *block, int i, uint8_t *dest, int line_size)
1892 {
1893     if (s->block_last_index[i] >= 0) {
1894         s->idsp.idct_add(dest, line_size, block);
1895     }
1896 }
1897
1898 static inline void add_dequant_dct(MpegEncContext *s,
1899                            int16_t *block, int i, uint8_t *dest, int line_size, int qscale)
1900 {
1901     if (s->block_last_index[i] >= 0) {
1902         s->dct_unquantize_inter(s, block, i, qscale);
1903
1904         s->idsp.idct_add(dest, line_size, block);
1905     }
1906 }
1907
1908 /**
1909  * Clean dc, ac, coded_block for the current non-intra MB.
1910  */
1911 void ff_clean_intra_table_entries(MpegEncContext *s)
1912 {
1913     int wrap = s->b8_stride;
1914     int xy = s->block_index[0];
1915
1916     s->dc_val[0][xy           ] =
1917     s->dc_val[0][xy + 1       ] =
1918     s->dc_val[0][xy     + wrap] =
1919     s->dc_val[0][xy + 1 + wrap] = 1024;
1920     /* ac pred */
1921     memset(s->ac_val[0][xy       ], 0, 32 * sizeof(int16_t));
1922     memset(s->ac_val[0][xy + wrap], 0, 32 * sizeof(int16_t));
1923     if (s->msmpeg4_version>=3) {
1924         s->coded_block[xy           ] =
1925         s->coded_block[xy + 1       ] =
1926         s->coded_block[xy     + wrap] =
1927         s->coded_block[xy + 1 + wrap] = 0;
1928     }
1929     /* chroma */
1930     wrap = s->mb_stride;
1931     xy = s->mb_x + s->mb_y * wrap;
1932     s->dc_val[1][xy] =
1933     s->dc_val[2][xy] = 1024;
1934     /* ac pred */
1935     memset(s->ac_val[1][xy], 0, 16 * sizeof(int16_t));
1936     memset(s->ac_val[2][xy], 0, 16 * sizeof(int16_t));
1937
1938     s->mbintra_table[xy]= 0;
1939 }
1940
1941 /* generic function called after a macroblock has been parsed by the
1942    decoder or after it has been encoded by the encoder.
1943
1944    Important variables used:
1945    s->mb_intra : true if intra macroblock
1946    s->mv_dir   : motion vector direction
1947    s->mv_type  : motion vector type
1948    s->mv       : motion vector
1949    s->interlaced_dct : true if interlaced dct used (mpeg2)
1950  */
1951 static av_always_inline
1952 void mpv_reconstruct_mb_internal(MpegEncContext *s, int16_t block[12][64],
1953                             int lowres_flag, int is_mpeg12)
1954 {
1955     const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
1956
1957     if (CONFIG_XVMC &&
1958         s->avctx->hwaccel && s->avctx->hwaccel->decode_mb) {
1959         s->avctx->hwaccel->decode_mb(s);//xvmc uses pblocks
1960         return;
1961     }
1962
1963     if(s->avctx->debug&FF_DEBUG_DCT_COEFF) {
1964        /* print DCT coefficients */
1965        int i,j;
1966        av_log(s->avctx, AV_LOG_DEBUG, "DCT coeffs of MB at %dx%d:\n", s->mb_x, s->mb_y);
1967        for(i=0; i<6; i++){
1968            for(j=0; j<64; j++){
1969                av_log(s->avctx, AV_LOG_DEBUG, "%5d",
1970                       block[i][s->idsp.idct_permutation[j]]);
1971            }
1972            av_log(s->avctx, AV_LOG_DEBUG, "\n");
1973        }
1974     }
1975
1976     s->current_picture.qscale_table[mb_xy] = s->qscale;
1977
1978     /* update DC predictors for P macroblocks */
1979     if (!s->mb_intra) {
1980         if (!is_mpeg12 && (s->h263_pred || s->h263_aic)) {
1981             if(s->mbintra_table[mb_xy])
1982                 ff_clean_intra_table_entries(s);
1983         } else {
1984             s->last_dc[0] =
1985             s->last_dc[1] =
1986             s->last_dc[2] = 128 << s->intra_dc_precision;
1987         }
1988     }
1989     else if (!is_mpeg12 && (s->h263_pred || s->h263_aic))
1990         s->mbintra_table[mb_xy]=1;
1991
1992     if ((s->avctx->flags & AV_CODEC_FLAG_PSNR) || s->frame_skip_threshold || s->frame_skip_factor ||
1993         !(s->encoding && (s->intra_only || s->pict_type == AV_PICTURE_TYPE_B) &&
1994           s->avctx->mb_decision != FF_MB_DECISION_RD)) { // FIXME precalc
1995         uint8_t *dest_y, *dest_cb, *dest_cr;
1996         int dct_linesize, dct_offset;
1997         op_pixels_func (*op_pix)[4];
1998         qpel_mc_func (*op_qpix)[16];
1999         const int linesize   = s->current_picture.f->linesize[0]; //not s->linesize as this would be wrong for field pics
2000         const int uvlinesize = s->current_picture.f->linesize[1];
2001         const int readable= s->pict_type != AV_PICTURE_TYPE_B || s->encoding || s->avctx->draw_horiz_band || lowres_flag;
2002         const int block_size= lowres_flag ? 8>>s->avctx->lowres : 8;
2003
2004         /* avoid copy if macroblock skipped in last frame too */
2005         /* skip only during decoding as we might trash the buffers during encoding a bit */
2006         if(!s->encoding){
2007             uint8_t *mbskip_ptr = &s->mbskip_table[mb_xy];
2008
2009             if (s->mb_skipped) {
2010                 s->mb_skipped= 0;
2011                 av_assert2(s->pict_type!=AV_PICTURE_TYPE_I);
2012                 *mbskip_ptr = 1;
2013             } else if(!s->current_picture.reference) {
2014                 *mbskip_ptr = 1;
2015             } else{
2016                 *mbskip_ptr = 0; /* not skipped */
2017             }
2018         }
2019
2020         dct_linesize = linesize << s->interlaced_dct;
2021         dct_offset   = s->interlaced_dct ? linesize : linesize * block_size;
2022
2023         if(readable){
2024             dest_y=  s->dest[0];
2025             dest_cb= s->dest[1];
2026             dest_cr= s->dest[2];
2027         }else{
2028             dest_y = s->sc.b_scratchpad;
2029             dest_cb= s->sc.b_scratchpad+16*linesize;
2030             dest_cr= s->sc.b_scratchpad+32*linesize;
2031         }
2032
2033         if (!s->mb_intra) {
2034             /* motion handling */
2035             /* decoding or more than one mb_type (MC was already done otherwise) */
2036             if(!s->encoding){
2037
2038                 if(HAVE_THREADS && s->avctx->active_thread_type&FF_THREAD_FRAME) {
2039                     if (s->mv_dir & MV_DIR_FORWARD) {
2040                         ff_thread_await_progress(&s->last_picture_ptr->tf,
2041                                                  lowest_referenced_row(s, 0),
2042                                                  0);
2043                     }
2044                     if (s->mv_dir & MV_DIR_BACKWARD) {
2045                         ff_thread_await_progress(&s->next_picture_ptr->tf,
2046                                                  lowest_referenced_row(s, 1),
2047                                                  0);
2048                     }
2049                 }
2050
2051                 if(lowres_flag){
2052                     h264_chroma_mc_func *op_pix = s->h264chroma.put_h264_chroma_pixels_tab;
2053
2054                     if (s->mv_dir & MV_DIR_FORWARD) {
2055                         MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix);
2056                         op_pix = s->h264chroma.avg_h264_chroma_pixels_tab;
2057                     }
2058                     if (s->mv_dir & MV_DIR_BACKWARD) {
2059                         MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix);
2060                     }
2061                 }else{
2062                     op_qpix = s->me.qpel_put;
2063                     if ((!s->no_rounding) || s->pict_type==AV_PICTURE_TYPE_B){
2064                         op_pix = s->hdsp.put_pixels_tab;
2065                     }else{
2066                         op_pix = s->hdsp.put_no_rnd_pixels_tab;
2067                     }
2068                     if (s->mv_dir & MV_DIR_FORWARD) {
2069                         ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix, op_qpix);
2070                         op_pix = s->hdsp.avg_pixels_tab;
2071                         op_qpix= s->me.qpel_avg;
2072                     }
2073                     if (s->mv_dir & MV_DIR_BACKWARD) {
2074                         ff_mpv_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix, op_qpix);
2075                     }
2076                 }
2077             }
2078
2079             /* skip dequant / idct if we are really late ;) */
2080             if(s->avctx->skip_idct){
2081                 if(  (s->avctx->skip_idct >= AVDISCARD_NONREF && s->pict_type == AV_PICTURE_TYPE_B)
2082                    ||(s->avctx->skip_idct >= AVDISCARD_NONKEY && s->pict_type != AV_PICTURE_TYPE_I)
2083                    || s->avctx->skip_idct >= AVDISCARD_ALL)
2084                     goto skip_idct;
2085             }
2086
2087             /* add dct residue */
2088             if(s->encoding || !(   s->msmpeg4_version || s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO
2089                                 || (s->codec_id==AV_CODEC_ID_MPEG4 && !s->mpeg_quant))){
2090                 add_dequant_dct(s, block[0], 0, dest_y                          , dct_linesize, s->qscale);
2091                 add_dequant_dct(s, block[1], 1, dest_y              + block_size, dct_linesize, s->qscale);
2092                 add_dequant_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize, s->qscale);
2093                 add_dequant_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
2094
2095                 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2096                     if (s->chroma_y_shift){
2097                         add_dequant_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
2098                         add_dequant_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
2099                     }else{
2100                         dct_linesize >>= 1;
2101                         dct_offset >>=1;
2102                         add_dequant_dct(s, block[4], 4, dest_cb,              dct_linesize, s->chroma_qscale);
2103                         add_dequant_dct(s, block[5], 5, dest_cr,              dct_linesize, s->chroma_qscale);
2104                         add_dequant_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
2105                         add_dequant_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
2106                     }
2107                 }
2108             } else if(is_mpeg12 || (s->codec_id != AV_CODEC_ID_WMV2)){
2109                 add_dct(s, block[0], 0, dest_y                          , dct_linesize);
2110                 add_dct(s, block[1], 1, dest_y              + block_size, dct_linesize);
2111                 add_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize);
2112                 add_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize);
2113
2114                 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2115                     if(s->chroma_y_shift){//Chroma420
2116                         add_dct(s, block[4], 4, dest_cb, uvlinesize);
2117                         add_dct(s, block[5], 5, dest_cr, uvlinesize);
2118                     }else{
2119                         //chroma422
2120                         dct_linesize = uvlinesize << s->interlaced_dct;
2121                         dct_offset   = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
2122
2123                         add_dct(s, block[4], 4, dest_cb, dct_linesize);
2124                         add_dct(s, block[5], 5, dest_cr, dct_linesize);
2125                         add_dct(s, block[6], 6, dest_cb+dct_offset, dct_linesize);
2126                         add_dct(s, block[7], 7, dest_cr+dct_offset, dct_linesize);
2127                         if(!s->chroma_x_shift){//Chroma444
2128                             add_dct(s, block[8], 8, dest_cb+block_size, dct_linesize);
2129                             add_dct(s, block[9], 9, dest_cr+block_size, dct_linesize);
2130                             add_dct(s, block[10], 10, dest_cb+block_size+dct_offset, dct_linesize);
2131                             add_dct(s, block[11], 11, dest_cr+block_size+dct_offset, dct_linesize);
2132                         }
2133                     }
2134                 }//fi gray
2135             }
2136             else if (CONFIG_WMV2_DECODER || CONFIG_WMV2_ENCODER) {
2137                 ff_wmv2_add_mb(s, block, dest_y, dest_cb, dest_cr);
2138             }
2139         } else {
2140             /* Only MPEG-4 Simple Studio Profile is supported in > 8-bit mode.
2141                TODO: Integrate 10-bit properly into mpegvideo.c so that ER works properly */
2142             if (s->avctx->bits_per_raw_sample > 8){
2143                 const int act_block_size = block_size * 2;
2144
2145                 if(s->dpcm_direction == 0) {
2146                     s->idsp.idct_put(dest_y,                           dct_linesize, (int16_t*)(*s->block32)[0]);
2147                     s->idsp.idct_put(dest_y              + act_block_size, dct_linesize, (int16_t*)(*s->block32)[1]);
2148                     s->idsp.idct_put(dest_y + dct_offset,              dct_linesize, (int16_t*)(*s->block32)[2]);
2149                     s->idsp.idct_put(dest_y + dct_offset + act_block_size, dct_linesize, (int16_t*)(*s->block32)[3]);
2150
2151                     dct_linesize = uvlinesize << s->interlaced_dct;
2152                     dct_offset   = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
2153
2154                     s->idsp.idct_put(dest_cb,              dct_linesize, (int16_t*)(*s->block32)[4]);
2155                     s->idsp.idct_put(dest_cr,              dct_linesize, (int16_t*)(*s->block32)[5]);
2156                     s->idsp.idct_put(dest_cb + dct_offset, dct_linesize, (int16_t*)(*s->block32)[6]);
2157                     s->idsp.idct_put(dest_cr + dct_offset, dct_linesize, (int16_t*)(*s->block32)[7]);
2158                     if(!s->chroma_x_shift){//Chroma444
2159                         s->idsp.idct_put(dest_cb + act_block_size,              dct_linesize, (int16_t*)(*s->block32)[8]);
2160                         s->idsp.idct_put(dest_cr + act_block_size,              dct_linesize, (int16_t*)(*s->block32)[9]);
2161                         s->idsp.idct_put(dest_cb + act_block_size + dct_offset, dct_linesize, (int16_t*)(*s->block32)[10]);
2162                         s->idsp.idct_put(dest_cr + act_block_size + dct_offset, dct_linesize, (int16_t*)(*s->block32)[11]);
2163                     }
2164                 } else if(s->dpcm_direction == 1) {
2165                     int i, w, h;
2166                     uint16_t *dest_pcm[3] = {(uint16_t*)dest_y, (uint16_t*)dest_cb, (uint16_t*)dest_cr};
2167                     int linesize[3] = {dct_linesize, uvlinesize, uvlinesize};
2168                     for(i = 0; i < 3; i++) {
2169                         int idx = 0;
2170                         int vsub = i ? s->chroma_y_shift : 0;
2171                         int hsub = i ? s->chroma_x_shift : 0;
2172                         for(h = 0; h < (16 >> vsub); h++){
2173                             for(w = 0; w < (16 >> hsub); w++)
2174                                 dest_pcm[i][w] = (*s->dpcm_macroblock)[i][idx++];
2175                             dest_pcm[i] += linesize[i] / 2;
2176                         }
2177                     }
2178                 } else if(s->dpcm_direction == -1) {
2179                     int i, w, h;
2180                     uint16_t *dest_pcm[3] = {(uint16_t*)dest_y, (uint16_t*)dest_cb, (uint16_t*)dest_cr};
2181                     int linesize[3] = {dct_linesize, uvlinesize, uvlinesize};
2182                     for(i = 0; i < 3; i++) {
2183                         int idx = 0;
2184                         int vsub = i ? s->chroma_y_shift : 0;
2185                         int hsub = i ? s->chroma_x_shift : 0;
2186                         dest_pcm[i] += (linesize[i] / 2) * ((16 >> vsub) - 1);
2187                         for(h = (16 >> vsub)-1; h >= 1; h--){
2188                             for(w = (16 >> hsub)-1; w >= 1; w--)
2189                                 dest_pcm[i][w] = (*s->dpcm_macroblock)[i][idx++];
2190                             dest_pcm[i] -= linesize[i] / 2;
2191                         }
2192                     }
2193                 }
2194             }
2195             /* dct only in intra block */
2196             else if(s->encoding || !(s->codec_id==AV_CODEC_ID_MPEG1VIDEO || s->codec_id==AV_CODEC_ID_MPEG2VIDEO)){
2197                 put_dct(s, block[0], 0, dest_y                          , dct_linesize, s->qscale);
2198                 put_dct(s, block[1], 1, dest_y              + block_size, dct_linesize, s->qscale);
2199                 put_dct(s, block[2], 2, dest_y + dct_offset             , dct_linesize, s->qscale);
2200                 put_dct(s, block[3], 3, dest_y + dct_offset + block_size, dct_linesize, s->qscale);
2201
2202                 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2203                     if(s->chroma_y_shift){
2204                         put_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
2205                         put_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
2206                     }else{
2207                         dct_offset >>=1;
2208                         dct_linesize >>=1;
2209                         put_dct(s, block[4], 4, dest_cb,              dct_linesize, s->chroma_qscale);
2210                         put_dct(s, block[5], 5, dest_cr,              dct_linesize, s->chroma_qscale);
2211                         put_dct(s, block[6], 6, dest_cb + dct_offset, dct_linesize, s->chroma_qscale);
2212                         put_dct(s, block[7], 7, dest_cr + dct_offset, dct_linesize, s->chroma_qscale);
2213                     }
2214                 }
2215             }else{
2216                 s->idsp.idct_put(dest_y,                           dct_linesize, block[0]);
2217                 s->idsp.idct_put(dest_y              + block_size, dct_linesize, block[1]);
2218                 s->idsp.idct_put(dest_y + dct_offset,              dct_linesize, block[2]);
2219                 s->idsp.idct_put(dest_y + dct_offset + block_size, dct_linesize, block[3]);
2220
2221                 if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2222                     if(s->chroma_y_shift){
2223                         s->idsp.idct_put(dest_cb, uvlinesize, block[4]);
2224                         s->idsp.idct_put(dest_cr, uvlinesize, block[5]);
2225                     }else{
2226
2227                         dct_linesize = uvlinesize << s->interlaced_dct;
2228                         dct_offset   = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
2229
2230                         s->idsp.idct_put(dest_cb,              dct_linesize, block[4]);
2231                         s->idsp.idct_put(dest_cr,              dct_linesize, block[5]);
2232                         s->idsp.idct_put(dest_cb + dct_offset, dct_linesize, block[6]);
2233                         s->idsp.idct_put(dest_cr + dct_offset, dct_linesize, block[7]);
2234                         if(!s->chroma_x_shift){//Chroma444
2235                             s->idsp.idct_put(dest_cb + block_size,              dct_linesize, block[8]);
2236                             s->idsp.idct_put(dest_cr + block_size,              dct_linesize, block[9]);
2237                             s->idsp.idct_put(dest_cb + block_size + dct_offset, dct_linesize, block[10]);
2238                             s->idsp.idct_put(dest_cr + block_size + dct_offset, dct_linesize, block[11]);
2239                         }
2240                     }
2241                 }//gray
2242             }
2243         }
2244 skip_idct:
2245         if(!readable){
2246             s->hdsp.put_pixels_tab[0][0](s->dest[0], dest_y ,   linesize,16);
2247             if (!CONFIG_GRAY || !(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {
2248                 s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[1], dest_cb, uvlinesize,16 >> s->chroma_y_shift);
2249                 s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[2], dest_cr, uvlinesize,16 >> s->chroma_y_shift);
2250             }
2251         }
2252     }
2253 }
2254
2255 void ff_mpv_reconstruct_mb(MpegEncContext *s, int16_t block[12][64])
2256 {
2257 #if !CONFIG_SMALL
2258     if(s->out_format == FMT_MPEG1) {
2259         if(s->avctx->lowres) mpv_reconstruct_mb_internal(s, block, 1, 1);
2260         else                 mpv_reconstruct_mb_internal(s, block, 0, 1);
2261     } else
2262 #endif
2263     if(s->avctx->lowres) mpv_reconstruct_mb_internal(s, block, 1, 0);
2264     else                  mpv_reconstruct_mb_internal(s, block, 0, 0);
2265 }
2266
2267 void ff_mpeg_draw_horiz_band(MpegEncContext *s, int y, int h)
2268 {
2269     ff_draw_horiz_band(s->avctx, s->current_picture_ptr->f,
2270                        s->last_picture_ptr ? s->last_picture_ptr->f : NULL, y, h, s->picture_structure,
2271                        s->first_field, s->low_delay);
2272 }
2273
2274 void ff_init_block_index(MpegEncContext *s){ //FIXME maybe rename
2275     const int linesize   = s->current_picture.f->linesize[0]; //not s->linesize as this would be wrong for field pics
2276     const int uvlinesize = s->current_picture.f->linesize[1];
2277     const int width_of_mb = (4 + (s->avctx->bits_per_raw_sample > 8)) - s->avctx->lowres;
2278     const int height_of_mb = 4 - s->avctx->lowres;
2279
2280     s->block_index[0]= s->b8_stride*(s->mb_y*2    ) - 2 + s->mb_x*2;
2281     s->block_index[1]= s->b8_stride*(s->mb_y*2    ) - 1 + s->mb_x*2;
2282     s->block_index[2]= s->b8_stride*(s->mb_y*2 + 1) - 2 + s->mb_x*2;
2283     s->block_index[3]= s->b8_stride*(s->mb_y*2 + 1) - 1 + s->mb_x*2;
2284     s->block_index[4]= s->mb_stride*(s->mb_y + 1)                + s->b8_stride*s->mb_height*2 + s->mb_x - 1;
2285     s->block_index[5]= s->mb_stride*(s->mb_y + s->mb_height + 2) + s->b8_stride*s->mb_height*2 + s->mb_x - 1;
2286     //block_index is not used by mpeg2, so it is not affected by chroma_format
2287
2288     s->dest[0] = s->current_picture.f->data[0] + (int)((s->mb_x - 1U) <<  width_of_mb);
2289     s->dest[1] = s->current_picture.f->data[1] + (int)((s->mb_x - 1U) << (width_of_mb - s->chroma_x_shift));
2290     s->dest[2] = s->current_picture.f->data[2] + (int)((s->mb_x - 1U) << (width_of_mb - s->chroma_x_shift));
2291
2292     if(!(s->pict_type==AV_PICTURE_TYPE_B && s->avctx->draw_horiz_band && s->picture_structure==PICT_FRAME))
2293     {
2294         if(s->picture_structure==PICT_FRAME){
2295         s->dest[0] += s->mb_y *   linesize << height_of_mb;
2296         s->dest[1] += s->mb_y * uvlinesize << (height_of_mb - s->chroma_y_shift);
2297         s->dest[2] += s->mb_y * uvlinesize << (height_of_mb - s->chroma_y_shift);
2298         }else{
2299             s->dest[0] += (s->mb_y>>1) *   linesize << height_of_mb;
2300             s->dest[1] += (s->mb_y>>1) * uvlinesize << (height_of_mb - s->chroma_y_shift);
2301             s->dest[2] += (s->mb_y>>1) * uvlinesize << (height_of_mb - s->chroma_y_shift);
2302             av_assert1((s->mb_y&1) == (s->picture_structure == PICT_BOTTOM_FIELD));
2303         }
2304     }
2305 }
2306
2307 void ff_mpeg_flush(AVCodecContext *avctx){
2308     int i;
2309     MpegEncContext *s = avctx->priv_data;
2310
2311     if (!s || !s->picture)
2312         return;
2313
2314     for (i = 0; i < MAX_PICTURE_COUNT; i++)
2315         ff_mpeg_unref_picture(s->avctx, &s->picture[i]);
2316     s->current_picture_ptr = s->last_picture_ptr = s->next_picture_ptr = NULL;
2317
2318     ff_mpeg_unref_picture(s->avctx, &s->current_picture);
2319     ff_mpeg_unref_picture(s->avctx, &s->last_picture);
2320     ff_mpeg_unref_picture(s->avctx, &s->next_picture);
2321
2322     s->mb_x= s->mb_y= 0;
2323     s->closed_gop= 0;
2324
2325     s->parse_context.state= -1;
2326     s->parse_context.frame_start_found= 0;
2327     s->parse_context.overread= 0;
2328     s->parse_context.overread_index= 0;
2329     s->parse_context.index= 0;
2330     s->parse_context.last_index= 0;
2331     s->bitstream_buffer_size=0;
2332     s->pp_time=0;
2333 }
2334
2335 /**
2336  * set qscale and update qscale dependent variables.
2337  */
2338 void ff_set_qscale(MpegEncContext * s, int qscale)
2339 {
2340     if (qscale < 1)
2341         qscale = 1;
2342     else if (qscale > 31)
2343         qscale = 31;
2344
2345     s->qscale = qscale;
2346     s->chroma_qscale= s->chroma_qscale_table[qscale];
2347
2348     s->y_dc_scale= s->y_dc_scale_table[ qscale ];
2349     s->c_dc_scale= s->c_dc_scale_table[ s->chroma_qscale ];
2350 }
2351
2352 void ff_mpv_report_decode_progress(MpegEncContext *s)
2353 {
2354     if (s->pict_type != AV_PICTURE_TYPE_B && !s->partitioned_frame && !s->er.error_occurred)
2355         ff_thread_report_progress(&s->current_picture_ptr->tf, s->mb_y, 0);
2356 }